Revert "[AMD] Fix DeepSeek V4 Pro c128 state tensor dtype mismatch error and c4_sparse_raw_indices attribute error in cuda graph phase"

[At1a8]

Reverts #27529, since it cause

• kernel perf regression issue noted here: [AMD] Fix DeepSeek V4 Pro c128 state tensor dtype mismatch error and c4_sparse_raw_indices attribute error in cuda graph phase #27529 (review)
• cuda graph launch error has been solved by [AMD] Add unified kv attention support in dpsk-v4 #27380, and it's logic also conflict with [AMD] Fix DeepSeek V4 Pro c128 state tensor dtype mismatch error and c4_sparse_raw_indices attribute error in cuda graph phase #27529 which will cause crush issue again.

---

CI States

Latest PR Test (Base): ❌ Run #27345325269
Latest PR Test (Extra): ❌ Run #27345324822

[gemini-code-assist]

Code Review

This pull request simplifies the DeepSeek-V4 JIT compression kernels (c128 and c4) by removing the BufFloat template parameter and unifying the KV buffer type with the input type (InFloat). This removes redundant template parameters, simplifies memory loading and writing routines, and defers float casting to the computation phase. The Python bindings and compressor layers are updated accordingly to remove the unused buffer dtype parameter and associated casting hotfixes. No review comments were provided for this pull request.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.